Analysis Report — banking

Generated on 2025-10-06T23:09:08.344006
Rows: 8000
Positive: 34.7% • Negative: 40.0% • Neutral: 25.3

Executive summary

Overall sentiment: mean VADER compound = 0.042 (std 0.387).
Top positive terms: deposit, transaction, account, like, successful, recent
Top negative terms: account, transaction, overdraft, like, fraud, inquire

Notable time points

Most positive period: 2025-08-13 (mean=0.086).
Most negative period: 2025-09-03 (mean=0.017).

Top Terms

Sentiment Over Time

Predictive terms (predicting negative sentiment)

The table below lists features (tokens) that are most predictive of the negative label based on a simple logistic model.
term coef
overdraft 4.275944
fraud 4.126616
denied 3.433058
erroneous 3.200109
disputed 3.154114
suspicious 2.465335
unauthorized 2.298348
ref 2.019261
ticket 0.772830
port 0.664804

Download CSV

Sentiment spikes

Detected spikes in rolling sentiment (z-score threshold). Review the rows to understand dates and severity.
timestamp sentiment_mean count rolling_mean rolling_std zscore spike
2025-07-09 0.043634 574 0.043634 0.000000 0.000000 False
2025-07-16 0.018188 628 0.030911 0.017993 -0.707107 False
2025-07-23 0.043290 624 0.035038 0.014593 0.565538 False
2025-07-30 0.039770 621 0.033750 0.013591 0.442970 False
2025-08-06 0.050619 630 0.044560 0.005535 1.094747 False
2025-08-13 0.085805 616 0.058731 0.024066 1.124985 False
2025-08-20 0.059123 623 0.065182 0.018359 -0.330033 False
2025-08-27 0.043011 628 0.062646 0.021613 -0.908483 False
2025-09-03 0.017131 611 0.039755 0.021185 -1.067950 False
2025-09-10 0.057415 634 0.039186 0.020413 0.893043 False

Download CSV

Topic — sentiment correlation

Correlation between LDA topic weights and VADER sentiment (positive values indicate topics associated with more positive sentiment).
topic keywords corr_with_sentiment
topic_0 account, like, inquire, saving, overdraft, fraud 0.341223
topic_1 report, writing, issue, suspicious, unauthorized, pending -0.040391
topic_2 correctly, recent, processed, fraud, overdraft, deposit -0.051491
topic_3 hold, need, transaction, high, value, erroneous -0.258176
topic_4 fund, insufficient, disputed, approved, deposit, overdraft -0.083368
topic_5 transaction, need, hold, urgent, successful, declined 0.015024

Download CSV

Clusters — top terms & samples

Clusters are produced by LSA embeddings + KMeans; top terms summarize each cluster and sample snippets give context.

Cluster top terms

Cluster 0: transaction, successful, urgent, pending, value, denied, high, declined, disputed, erroneous

Cluster 1: saving, inquire, like, account, overdraft, fraud, deposit, loan, payment, statement

Cluster 2: correctly, processed, recent, fraud, overdraft, deposit, mortgage, payment, balance, rate

Cluster 3: need, hold, urgent, erroneous, unauthorized, denied, flagged, successful, disputed, high

Cluster 4: issue, writing, report, suspicious, unauthorized, pending, successful, erroneous, declined, high

Cluster 5: insufficient, fund, disputed, approved, withdrew, charged, transferred, flagged, denied, resolved

Cluster samples

Cluster 0:
urgent transaction 2241.65 fraud investigated
like urgent transaction 746.75 account investigated
unauthorized transaction 1974.07 interest rate flagged

Cluster 1:
would like inquire deposit saving account
would like inquire deposit saving account
would like inquire loan saving account

Cluster 2:
recent payment 1046.1 processed correctly
ref-9617 recent mortgage 3830.51 processed correctly
recent statement 4242.36 processed correctly

Cluster 3:
suspicious hold interest rate need disputed
ref-7751 pending hold fraud need denied
high-value hold payment need withdrew

Cluster 4:
writing report suspicious issue fraud
writing report suspicious issue account
writing report high-value issue deposit

Cluster 5:
statement deposited due insufficient fund
statement disputed due insufficient fund
loan flagged due insufficient fund

Download cluster top terms • Download cluster samples